Erik Kusch
,
PhD Student
Department
of Biology
Section for Ec
oinformatics &
Biodiversity
Center for
Biodiversi
ty Dy
namics
in a Changing World (BIOCHANGE)
Aarhus University
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
1
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
2
Models need to straddle the lin
e betw
een
being too
simplistic and being overly
complex
.
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
3
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
4
Models ought to be regul
arised (tuned for
desired comple
xity-accuracy)!
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
5
W
e can compare models w
ithout
know
ing
the absolute truth.
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
6
Dif
ference
in
log-score
between
training and
test data
reveals
overfitting
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
7
Regularised priors f
air w
orse
in
-
sample, but better out-
of
-sample.
Regula
risation
ef
fects diminish w
ith
increasing sample sizes.
05/02/2021
[Study
Group]
Bayesian
Statistics
w
ith the Rethinking
M
aterial
8
This is the
model we would want
to
select
because i
t shows the
treatment
effect
Model Selection can lead to mis-
identification of causal pathway
s.